A new framework for creating original ideas with AI
Listen to this article:
“Give me 20 headlines.” “Create 30 research questions.” “Generate 50 business ideas.” Beneath these requests lies an intuitive assumption about large language models: that originality emerges automatically from quantity. If an AI can generate ideas faster than humans, then perhaps creativity becomes a problem of volume. Yet anyone who has brainstormed with ChatGPT or another LLM has quickly discovered the limitations of this approach. The outputs are often confident and polished, but also slightly repetitive and just off the mark. An important prompting challenge, then, is how to think with AI in ways that intelligently boost novelty. The paper “Ideation with Generative AI—in Consumer Research and Beyond,” published in the Journal of Consumer Research and co-written by HBS AI Institute Julian De Freitas, offers a new framework for understanding both what LLMs bring to brainstorming and where they fall short. The result is a practical roadmap for supercharging creativity.
Key Insight: The Engines of AI Creativity
“We argue that productivity is loosely analogous to the ‘persistence path,’ and semantic breadth to the ‘flexibility path’ to creativity in human psychology.” [1]
The researchers draw a useful parallel between how LLMs generate ideas and how human creativity works. Humans reach creative breakthroughs through two distinct psychological pathways: persistence (deep, focused exploration within a constrained space) and flexibility (drawing unexpected connections across distant domains). LLMs have functional analogues for both: productivity (the sheer speed and volume with which LLMs generate outputs) and semantic breadth (the ability to link highly distant concepts). While an individual’s originality spikes when using an LLM (thanks to these capabilities), collective diversity plummets. Because models predict highly probable word sequences from their training data, independent users default to similar, co-occurring patterns, which flattens group variety. To overcome this limitation, the authors recommend several techniques including prompt variation (e.g. by employing personas), hybrid prompting (combining idea pools from distinct prompts), chain-of-thought prompting (guiding the model through ordered, step-by-step creative refinements), and temperature parameters (strategically using low, medium, and high ‘temperature’ prompt variations).
Key Insight: Four Roles for Human-AI Co-Creation
“[T]o utilize LLMs for big ideas, humans may have to assume the role of key ideator in the co-creation process.” [2]
One of the paper’s most actionable contributions is a framework of four metaphorical roles that LLMs can play in ideation: designer, writer, interviewer, and actor. The roles differ in where the creative lead sits: as a designer or writer, the LLM helps generate or shape the output; as interviewer or actor, the LLM is used to draw better ideas out of the human. For the designer, have the LLM act like a research setup assistant. For example, with the authors’ use case of consumer research, the LLM might create test ads, product descriptions, and scenarios, and then look for hidden differences that could distort the results. In the writer role, the user prompts the LLM to act like a communication coach, taking ideas that may be rough, abstract, or overly technical and expressing them in ways that are clearer, more concrete, and more persuasive. For the interviewer, the user prompts the LLM to ask them questions rather than generate answers. The goal is to use the LLM as a guided brainstorming partner, pushing the user to challenge assumptions, draw on their own experiences, or connect ideas that would not normally sit together. For the actor role, the user prompts the LLM to roleplay a specific consumer, customer, or stakeholder so that the user can ask it exploratory interview questions. Importantly, the actor role is for inspiration and preparation, not evidence. Since the LLM is not a real consumer, its responses should not replace real customer interviews, surveys, or behavioral data.
Why This Matters
The authors explain that current LLMs excel primarily at producing incremental variations (“small ideas”) rather than paradigm-shifting breakthroughs (“big ideas”), so organizations that completely replace human-driven work with AI-generated output face an unseen ceiling on true innovation. Leaders and executives making decisions about product innovation, customer research, marketing strategy, or organizational change, should encourage their employees to use LLMs as configurable collaborators. The companies that benefit most will not simply produce more ideas, they’ll keep their employees engaged, test assumptions, and find real insights that merit investment.
Bonus
Want to go deeper? The original paper includes 10 concrete guidelines for utilizing LLMs in ideation.
References
[1] De Freitas, Julian, Gideon Nave, and Stefano Puntoni, “Ideation with Generative AI—in Consumer Research and Beyond,” Journal of Consumer Research, Volume 52, Issue 1 (June 2025): 19.
[2] De Freitas et al., “Ideation with Generative AI,” 23.
Meet the Authors
Julian De Freitas is an Assistant Professor of Business Administration in the Marketing Unit and Director of the Ethical Intelligence Lab at Harvard Business School, and Associate HBS AI Institute. His work sits at the nexus of AI, consumer psychology, and ethics.
Gideon Nave is Carlos and Rosa de la Cruz Associate Professor at The Wharton School of the University of Pennsylvania.
Stefano Puntoni is Sebastian S. Kresge Professor of Marketing at The Wharton School of the University of Pennsylvania.
Watch a video version of the Insight Article here.